Diabetic Retinopathy Severity Identification Using 3D Dual-Domain Attention Approach

Abstract

The diabetic retinopathy (DR) is one of prominent reason of visual impairment among the people around the globe suffers from diabetes. Early and timely diagnosis of this problem can minimise the risk of proliferated diabetic retinopathy. Diabetes is caused by persistently high blood glucose levels, which leads to blood vessel aggravations and vision loss. Therefore, it becomes important to classify DR stages. An automated system for this purpose contains several phases like identification and classification of DR stages in fundus images. Deep learning techniques based on extraction of features and automatic extraction of features with a hybrid network have been presented for diabetic retinopathy detection. This method effectively identify diabetic retinopathy identification from the chest region by using the 3D Dual-Domain Attention Approach. The dual-domain attention module propised learns local and global information in spatial and context domains from encoding feature maps in Unet. Our attention module generates refined feature maps from the enlarged reception field at every stage by attention mechanisms and residual learning to focus on complex tumor regions. Experimental results show that the proposed network can identify the DR stages with high accuracy. The proposed method attains an F1-score of 91.34%, precision of 92.34%, accuracy of 98.65%, on the healthy retina, stage 1, stage 2, and stage 3 fundus images. Compared with other models, our proposed network achieves comparable performance.